Skip to content
All library documents

Evaluating Long-Volatility Strategies Through Drawdowns and Waiting Times

Article Quant Q&A · Author: Richard Harb
Read the original

Summary

The document considers how to assess long-volatility strategies with frequent small losses and occasional large gains, where mean return, Sharpe ratio, and hit rate may not capture the payoff pattern. The response recommends beginning with a long history of time-series data, then examining maximum drawdown and the distribution of drawdowns. It also highlights the average time spent waiting for large gains, since losses and the time required to realize a payoff can affect whether a trader can maintain the position.

The response retains Sharpe ratio as a relevant measure and frames drawdown as a practical concern because losses may create funding pressure or force an early exit. It suggests combining these strategies with other approaches that can offset ongoing losses or diversifying across independent event-driven strategies. These are practical evaluation considerations, not a complete statistical testing framework: the document does not specify a null model, formal tail tests, or methods to distinguish skill from luck. Its advice depends on having enough historical observations and on realistic assumptions about capital and strategy independence.

Key ideas

  • Long-volatility strategies can have negative average returns despite occasional large gains.
  • Maximum drawdown and the distribution of drawdowns help describe the losses a trader must withstand.
  • The time spent waiting for large gains is a relevant measure of strategy experience.
  • A long historical record is needed to assess rare payoff events.
  • Combining strategies may reduce the impact of persistent losses if their returns are sufficiently independent.

Tags

VolatilityBacktestingRisk managementPortfolio constructionStatistics
Full text
# How should long-volatility strategies be statistically evaluated beyond mean returns?


# How should long-volatility strategies be statistically evaluated beyond mean returns?












I am looking for a rigorous framework to test long-volatility strategies, specifically strategies with convex, asymmetric payoffs (many small losses, few large gains).

My concern is that most standard backtesting approaches (mean return, Sharpe ratio, hit rate) seem inappropriate for long-vol strategies, because:

Expected mean returns are often negative due to carry / decay Performance is driven by rare tail events, not average behavior Evaluating such strategies via average or median returns systematically rejects valid convex edges

My main questions: 1.What are the correct statistical metrics to evaluate long-vol strategies? (Tail probabilities? Quantile shifts? Distributional tests (KS, CvM)?)

What is the correct null model? (Random timestamps? Vol-regime matched random samples?)

How do practitioners distinguish skill vs luck when only a few outliers drive performance?

Are there established references (papers or books) that explicitly address the evaluation of convex / long-vol strategies, rather than directional or carry strategies?

## Answer by lehalle (score 2)

https://quant.stackexchange.com/a/85367

First you will need data, a long history of time series to get enough statistics to draw conclusions on your strategy.

Then I understand you are puzzled by the fact that the average does not reflects a stochastic process that has frequent small losses and rare intense gains. From a practical and industrial perspective a first perception of this property is the Maximum Draw Down (MDD): for how long will you wait, and more importantly how much money you will have to accept to loose, before seeing a gain?

This should be your focus because it means the margin calls you will have to face are proportional to this MMD, so if you cannot face them, you will unwind your position before any gain.

Besides the Sharpe ratio (that is nevertheless important), you can study the distribution of these MDDs, and the average waiting time before high gains.

In practice, such strategies are used in combination with a lot of others that will form a pocket of strategies with a not too painful max draw down: either you find a statistical arbitrage, market neutral strategy that reimburse this "bleeding", either you have access to a large number of independent "event-like strategies" to get the Central Limit Theorem averaging them.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.